databuff

DataBuff is an open-source monitoring tool for software teams that run multiple interconnected services, often called microservices. It collects data about how your services behave, how fast they respond, where errors occur, and how they call each other, then uses an AI assistant to help you investigate problems using plain English questions.

The monitoring side uses a standard called OpenTelemetry, which many programming frameworks already support. Once your services send data to DataBuff, it builds a visual map of all your services and their connections, tracks request rates and error counts, and records detailed traces of individual requests as they travel through your system.

The AI side is designed to query that monitoring data directly rather than through a generic chat interface added on top. You can ask questions like "why is service X slow?" and it will look at the actual traces, metrics, and alerts to give you an answer based on real data. For complex problems it can coordinate multiple AI agents, each investigating a different aspect of the issue at the same time.

DataBuff also exposes a connection point called MCP, which lets external AI tools like Cursor and Claude plug in and query your monitoring data directly. The README is primarily in Chinese and links to a live demo site.

Installation requires Docker and Docker Compose. A one-line install script sets up the full platform in about 5 minutes, after which you visit the web interface at port 27403 and enter an AI model API key to enable the AI features. Kubernetes installation is also available. The system runs on three components: an ingest layer, Apache Doris as the data store, and a web frontend.